ART: Adaptive Relational Transformer for Pedestrian Trajectory Prediction with Temporal-Aware Relations

2026-04-04

AI Summary

Existing approaches to pedestrian trajectory prediction often struggle with either computational redundancy or insufficient representational capacity when modeling dynamic and diverse interpersonal interactions. This work proposes the Adaptive Relation Transformer (ART), which introduces a novel Temporal-Aware Relation Graph (TARG) to explicitly capture the temporal evolution of social interactions and incorporates an Adaptive Interaction Pruning (AIP) mechanism to dynamically eliminate redundant computations. By jointly optimizing accuracy and efficiency, ART achieves state-of-the-art performance on both the ETH/UCY and NBA benchmarks, delivering superior prediction accuracy while significantly improving computational efficiency—all without compromising expressive power.

Technology Category

Humans and AI: Human-Aware Planning and Behavior PredictionReasoning under Uncertainty: Relational Probabilistic ModelsPlanning, Routing, and Scheduling: Optimization of Spatio-temporal Systems

Application Category

Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphsSearch and Retrieval-Augmented AI: Agentic searchUser Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Accurate prediction of real-world pedestrian trajectories is crucial for a wide range of robot-related applications. Recent approaches typically adopt graph-based or transformer-based frameworks to model interactions. Despite their effectiveness, these methods either introduce unnecessary computational overhead or struggle to represent the diverse and time-varying characteristics of human interactions. In this work, we present an Adaptive Relational Transformer (ART), which introduces a Temporal-Aware Relation Graph (TARG) to explicitly capture the evolution of pairwise interactions and an Adaptive Interaction Pruning (AIP) mechanism to reduce redundant computations efficiently. Extensive evaluations on ETH/UCY and NBA benchmarks show that ART delivers state-of-the-art accuracy with high computational efficiency.

Problem

• pedestrian trajectory prediction
• human interactions
• temporal-aware relations
• computational overhead
• interaction modeling

Innovation

• Adaptive Relational Transformer
• Temporal-Aware Relation Graph
• Adaptive Interaction Pruning
• trajectory prediction
• computational efficiency